1. Field of the Invention
The present invention relates to a technique required to control tone characteristics by applying different gains to respective portions of an input image.
2. Description of the Related Art
Conventionally, a digital dodging process for controlling tone characteristics by applying different gains to respective portions of an input image has been proposed. For example, this dodging process is applied to a so-called backlight scene in which the brightness of a main object to be shot is considerably lower than that of a background. In the backlight scene, when the brightness of an entire image is controlled to an appropriate level, the main object is generally shot darkly. Therefore, using luminance-dependent gains intended to apply a relatively large gain to a dark portion, the darkly shot main object is controlled to an appropriate brightness. The aforementioned process is the basic concept of the dodging process.
Furthermore, as an input luminance upon calculation of luminance-dependent gains, a low-frequency image generated from luminance components of an input image is used. This is because when an input image is used intact as an input luminance used to calculate luminance-dependent gains, a large gain is unwantedly applied to every dark object portions and black object portions in an image, and a contrast of an image after a gain process is considerably impaired. Therefore, using the low-frequency image, the influence of local dark portions in an image is reduced, and gain sensitivity to luminance changes of an image lowers, thus executing the dodging process while maintaining a resolution of an output image.
However, when the dodging process is executed using the low-frequency image, a pseudo edge is generated near edge portions between a low-luminance region and high-luminance region, as shown in FIG. 25. In the example shown in FIG. 25, a small blur signal and large blur signal as low-frequency components of a source signal are used in addition to the source signal. When a gain table has a pattern which is decreased largely relatively at a low-luminance side, as shown in FIG. 25, a gain converted from the blur signal is largely decreased on the low-luminance region side of an edge portion. This is because a value of the blur signal of the edge portion becomes large due to the influence of the high-luminance region side. Therefore, a black-rimmed pseudo edge is generated, as shown in FIG. 25. This phenomenon occurs dominantly as a range of a gain to be multiplied (a difference between GAIN_MAX and GAIN_MIN) becomes larger.
To solve the aforementioned problems, for example, techniques described in Japanese Patent Laid-Open Nos. 2010-244360 and 2009-272983 have been proposed.
In Japanese Patent Laid-Open No. 2010-244360, gains are calculated from a plurality of luminance images having different frequency bands. Then, when a gain calculated from a relatively low-frequency image exceeds that calculated from a relatively high-frequency image, the contribution of the gain calculated from the relatively low-frequency image is controlled to be reduced. More specifically, a use ratio of the gain calculated from the high-frequency image is increased.
In Japanese Patent Laid-Open No. 2009-272983, images having different frequency bands are used as in Japanese Patent Laid-Open No. 2010-244360, an image is divided into regions based on luminance values of a relatively high-frequency image, and a gain value calculated from a relatively low-frequency image is corrected by a region division result.
However, Japanese Patent Laid-Open No. 2010-244360 merely describes a measure against a case in which a gain calculated from a low-frequency image becomes relatively large, that is, a measure against pseudo tones like a white-rimmed high-luminance region, and cannot cope with pseudo tones of a low-luminance region. Also, when a gain to be multiplied with the low-luminance region becomes large, the method of Japanese Patent Laid-Open No. 2010-244360 cannot provide a sufficient effect, and pseudo tones are left unremoved.
Japanese Patent Laid-Open No. 2009-272983 does not describe a measure against a case in which gains having largely different gains are mixed in a single region, as shown in FIG. 25, and a gain to be re-calculated in that region is influenced by the largely different gains.